Hugging Face Papers Trend Monitor

Track fresh AI research from Hugging Face Daily Papers and detect which papers are gaining momentum instead of repeatedly exporting the same static feed.

What this Actor gives you

Each delivered paper can include:

• paper ID, title, abstract, authors and claimed Hugging Face author profiles
• publication and Daily Papers submission timestamps
• Hugging Face upvotes and comment count
• current feed rank
• linked GitHub repository, GitHub stars and project page when Hugging Face provides them
• organization and submitter metadata when available
• Hugging Face AI summary and AI keywords when available
• direct Hugging Face and arXiv links
• persistent change history under your own stateKey

The source is the public Hugging Face Daily Papers API.

The useful part: momentum tracking

On later runs with the same stateKey, the Actor compares the current paper snapshot with the previous one and adds:

• previousUpvotes / upvotesDelta
• previousRank / rankDelta
• previousComments / commentsDelta
• previousGithubStars / githubStarsDelta
• changeType: NEW, CHANGED, or UNCHANGED
• firstSeenAt and lastSeenAt

A positive rankDelta means the paper moved up in the feed (for example rank 20 → rank 5 gives rankDelta: 15). On the first observation, previous values and deltas are null.

This makes scheduled runs useful for research intelligence, newsletters, AI trend dashboards, RAG feeds, competitor/lab monitoring, and agent workflows.

Input examples

Current trending papers

{
  "sort": "trending",
  "limit": 50,
  "emitMode": "all"
}

Only new or changed papers about agents

{
  "keyword": "agents",
  "sort": "trending",
  "minUpvotes": 5,
  "emitMode": "new_or_changed",
  "stateKey": "agents-watch"
}

Scan a week

{
  "week": "2026-W39",
  "keyword": "vision language",
  "limit": 100,
  "maxPages": 5
}

You can use exactly one period selector at a time: dates, week, or month. If none is supplied, the latest feed is used. submitter can additionally restrict the Daily Papers feed to one Hugging Face submitter.

Search and filters

keyword uses case-insensitive AND matching across the available title, abstract, AI summary, authors, Hugging Face author usernames, AI keywords, organization, GitHub URL, and project page. minUpvotes can remove low-signal papers before delivery.

maxPages controls how deeply the Actor scans the upstream feed when filters are selective. Hugging Face currently allows up to 100 records per source page.

Delivery modes

• all — deliver every matching current paper
• new — deliver papers not previously seen under this stateKey
• changed — deliver only papers whose tracked momentum fields changed
• new_or_changed — deliver new papers plus changed existing papers

The older onlyNew input remains supported for backward compatibility. If it is enabled while emitMode is all, it behaves like emitMode: new.

The Actor stores up to 10,000 paper snapshots per monitoring key. Use a different stateKey for independent watchlists. Avoid concurrent runs with the same key because they share monitoring state.

Output

Typical fields include:

id, title, abstract, aiSummary, aiKeywords, authors, authorProfiles, publishedAt, submittedOnDailyAt, submittedBy, organization, upvotes, numComments, rank, githubRepo, githubStars, projectPage, changeType, delta fields, paperUrl, arxivUrl, firstSeenAt, lastSeenAt, and collectedAt.

Hugging Face does not populate every enrichment field for every paper, so GitHub/project/AI metadata can be empty.

Pricing

The PPE event delivered-paper costs $0.003 per delivered row ($3 / 1,000 delivered papers). Rows suppressed by emitMode are not delivered and do not trigger the PPE event. Platform compute/storage can still be consumed by a run even when no rows are delivered.

RUN_SUMMARY reports fetched, delivered, new, changed and unchanged counts plus the active delivery mode.

Reliability

The client uses bounded retries for transient HTTP 429 and 5xx responses and validates input limits before calling Hugging Face. Results still depend on the upstream public API and its current metadata.

This Actor is independent of and not endorsed by Hugging Face or arXiv.